The absolute error of a proposed time-series forecasting model divided by the absolute error of the random walk (no-change) model. The RAE is similar to Theil’s U2. The RAE can be averaged by taking a geometric mean (because the data are ratios) to get the Geometric Mean Relative Absolute Error (GMRAE). If outliers are expected, the GMRAE should be trimmed or the Median RAE (MdRAE) should be used. The GMRAE is used for calibrating models, and the MdRAE is used for comparing models. Armstrong (2001d) discusses the use of the RAE in comparing forecasts from different methods.

- Armstrong, J. S. (2001d), “Evaluating forecasting
methods,” in J. S. Armstrong (ed.),
*Principles of Forecasting.*Norwell, MA: Kluwer Academic Press.